Amir Afrashteh (High Radius): Agentic AI in treasury: A smarter way to forecast, manage, and move cash

Modern treasury is under pressure from real-time liquidity decisions to rising payment volumes and scattered data across systems. Traditional automation helps, but it often stops at task-level efficiency. What treasury teams now need is decision-level intelligence one that doesn’t just process information, but understands context and acts on it. Whether it’s anticipating a shortfall, rebalancing idle cash, mitigating FX risk, or learning from customer payment patterns, intelligence must be adaptive. This is where Agentic AI offers a shift bringing together intelligent agents that not only automate but reason and act in coordination to deliver meaningful outcomes across forecasting, cash management, and payments

How is treasury evolving from “automated” to “intelligent”—and what role does Agentic AI play?

Traditional systems follow fixed deterministic logic “if this, then that.” They work until exceptions occur or data changes. Agentic AI introduces a shift: agents that can interpret data, apply logic, and interact with one another to fulfill broader treasury goals. Think of it as going from spreadsheets and static rules to software agents that learn and adapt. In forecasting, for example, agents can analyze bank data and transaction patterns and intelligently choose the best-fit model to generate daily forecasts updating projections as new inputs flow in. This not only reduces manual intervention but also improves forecast quality and agility over time. Some treasury teams have reported short-term forecast accuracy improvements up to 95%, while reducing manual effort by nearly 80%. In practical terms, even a modest 1% improvement in forecast accuracy can unlock over $ 3M in working capital freeing up funds that would otherwise sit idle due to conservative cash buffers. This isn’t just about speed it’s about making confident, real-time decisions without compromising control.

How does Agentic AI bring intelligence and coordination to routine treasury operations?

The real strength of Agentic AI lies in how agents coordinate across workflows. In cash management, agents can monitor account balances in real time, flag idle cash, and even recommend internal transfers before shortfalls arise. Agents can also track bank cut-off times and suggest the most cost-effective routing options. Cash forecasting is on the sweet spot for Agentic AI where the agent can analyze the data, select the best fit model by running hundreds of combinations of forecasting models and parameters to drive the highest forecast accuracy given the dataset. The combination of automation and high forecast accuracy unlocks great business value across treasury organizations. Payments, though more structured, benefit equally from orchestration. Intelligent agents can process over 90% of vendor payments end-to-end extracting ERP files, generating bank-ready outputs, updating payment status across systems, and settling internal transfers or investment-related cash flows through integration with cash and debt modules. Fraud detection and compliance is a great example of Agentic AI where the agent analyses payment patterns and flags potential fraudulent and non-compliant payments. When exceptions or FX-related complexities emerge say, reconciling a blocked cross-border payment or navigating dual-approval rules assisted agents step in, keeping human oversight exactly where it’s needed. This hybrid model has delivered up to 30% savings in processing costs not just from automation, but from smarter exception handling, better routing choices, and improved control across fragmented banking networks